Action Synthesis Using Body Segmentation

Recognizing action is an important part of a video surveillance and video retrieval system, in which motion information extracted from video is useful. This paper proposes an action synthesis method, in which Efros et al.’s action recognition method, Chen et al.’s body segmentation method, and Fujiyosi et al.’s method are combined. Chen et al.’s method and Fujiyosi et al.’s method are used as preprocessing of action synthesis. Experimental results with a number of test sequences show that the proposed method works efficiently for human action synthesis.